Research map / Vision-language and multimodal / Human and embodied modeling
Decoupled Attention Network for Text Recognition
Tianwei Wang, Yuanzhi Zhu, Lianwen Jin, Canjie Luo, Xiaoxue Chen, Yaqiang Wu, Qianying Wang, Mingxiang Cai
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Replace recurrent attention alignment with visual-feature-only CAM and a decoupled decoder | Text recognition with the DAN feature encoder · IAM | CER: 9.9 -> 6.4 | Improved |
| Replace recurrent attention alignment with visual-feature-only CAM and a decoupled decoder | Text recognition with the DAN feature encoder · RIMES | WER: 9.1 -> 8.9 | Improved |
| Use visual-feature-only alignment instead of recurrent attention alignment | DAN on handwritten text · IAM | CER: The paper reports a 1.5% CER improvement over previous state-of-the-art. | Improved |
| Set CAM output length to 200 instead of 150 (ablation) | DAN · IAM | CER: 6.4 -> 6.3 | No clear effect |
| Set CAM output length to 250 instead of 150 (ablation) | DAN · IAM | CER: 6.4 -> 6.4 | No clear effect |
| Reduce CAM depth L (ablation) | DAN · IAM | Recognition performance | Hurt |
| Use 2D rather than 1D DAN | DAN · IIIT5K | accuracy: 93.3 -> 94.3 | Improved |
| Use 2D rather than 1D DAN | DAN · IC03 | accuracy: 95.2 -> 95.0 | Hurt |
| Use 2D rather than 1D DAN | DAN · IC13 | accuracy: 94.2 -> 93.9 | Hurt |
| Use 2D rather than 1D DAN | DAN · SVT-P | accuracy: 76.8 -> 80.0 | Improved |
| Use 2D rather than 1D DAN | DAN · CUTE80 | accuracy: 80.6 -> 84.4 | Improved |
| Use 2D rather than 1D DAN | DAN · IC15 | accuracy: 71.8 -> 74.5 | Improved |
Closest papers by meaning
- DAN: A Segmentation-Free Document Attention Network for Handwritten Document Recognition (TPAMI 2023 · similarity 0.90)
- An End-to-End TextSpotter With Explicit Alignment and Attention (CVPR 2018 · similarity 0.88)
- End-to-End Handwritten Paragraph Text Recognition Using a Vertical Attention Network (TPAMI 2022 · similarity 0.87)
- Focusing Attention: Towards Accurate Text Recognition in Natural Images (ICCV 2017 · similarity 0.87)
- Joint Visual Semantic Reasoning: Multi-Stage Decoder for Text Recognition (ICCV 2021 · similarity 0.86)
- RobustScanner: Dynamically Enhancing Positional Clues for Robust Text Recognition (ECCV 2020 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
